Monolith vs Microservices — a practical guide for startups and small teams

Published on 2025-09-25 21:13

TL;DR: Microservices are powerful but costly. For small teams (≈5 engineers) or early-stage startups, start with a modular monolith: design clear module boundaries, keep deployments simple, and extract services only when the operational and business benefits clearly outweigh the added cost.

Why this matters

Too often the conversation starts with "microservices" as if they were a default good. Microservices solve particular problems — team autonomy, independent scaling, and isolation — but they also introduce real complexity: more infra, more DevOps, distributed failures, harder testing, and extra hiring pressure. That complexity has a real cost, especially for small teams and early-stage products.

This post shows practical trade-offs, concrete patterns (including the modular monolith), a migration strategy, and a checklist to help you decide.

Quick definitions (practical, not academic)

Monolith

A single deployable application (one codebase, one runtime/deployment). Note: a monolith can still be well-structured and modular internally — it’s the deployment unit that’s “single.”

Microservices

An architecture composed of multiple independently deployable services. Each service often owns its own codebase, runtime, and (sometimes) datastore. Services communicate over the network via APIs or messaging.

Modular monolith

A single deployable application that is organized as discrete modules (bounded contexts) with clear interfaces and low coupling. It aims for the best of both worlds: simple deployment + modular boundaries that make future extraction easier.

Practical pros & cons

Monolith (practical lens)

Pros

Simple to deploy, run, and debug.
Single CI/CD pipeline: easier release process.
Transactions and data consistency are straightforward.
Lower initial infrastructure and tooling cost.
Faster feature iteration for small teams.

Cons

Can become a big-ball-of-mud without discipline.
Deploys touch everything (risk of blast radius).
Scaling a single process can be inefficient if different parts have different demands.
Can slow teams if the codebase grows and ownership is unclear.

Microservices

Pros

Independent deploys and release cadence per service.
Teams can own services end-to-end (code + infra + data).
Independent scaling tailored to each service’s needs.
Fault isolation (a failure in one service can be contained).

Cons

Operational complexity: many CI pipelines, infra, monitoring.
Distributed systems problems: network latency, retries, partial failures.
More expensive: more hosts/containers, more monitoring, more ops time.
Data consistency becomes harder (eventual consistency, saga patterns).
Debugging and tracing require instrumentation and tooling.
More contracts to design and maintain (APIs + versioning).

Costs to consider (beyond just money)

Engineering time — designing, implementing, testing cross-service contracts, and writing integration tests.
DevOps / SRE time — building pipelines, setting up orchestration, autoscaling, monitoring, runbooks.
Infrastructure costs — more instances/containers, load balancers, message brokers, managed DBs.
Observability tooling — distributed tracing, centralized logs, metrics & alerts.
Hiring & staffing — need or desire for SRE, platform engineers, or senior engineers who can own distributed systems.
Cognitive load — more concepts to train new hires on (eventual consistency, service meshes, retries/backoff strategies).

Team size & practical guidance

Solo developer – 3 people: Monolith or modular monolith. Simplicity and speed matter.
~4–10 people: Modular monolith is still preferable in most cases. Start modular; extract only when a boundary stabilizes.
>10 people or multiple cross-functional product teams: Consider microservices if organizational boundaries exist and you can support the operational overhead.

Important: team structure often drives architecture. If you have independent teams with separate owners and SLAs, microservices may make sense. If everyone pairs on the same features, a monolith is faster.

What a modular monolith looks like (concrete)

A modular monolith is built as a single deployable app but split into modules following bounded contexts. Each module:

Has a clear responsibility (e.g., Billing, Users, Orders).
Exposes a clear internal API/interface.
Owns its models and business logic.
Minimizes direct access to other modules’ internals.

Example folder layout (server-side)

/src
  /modules
    /users
      /models
      /services
      /api    # controllers / handlers
      index.ts
    /orders
      /models
      /services
      /api
      index.ts
    /billing
      ...
  /shared
    /database
    /auth
    /logger
  /infrastructure
    /jobs
    /background
  app.ts

Communication patterns inside a modular monolith

Synchronous internal method/API calls between modules (function calls or module interfaces).
In-process event bus for decoupling (publish/subscribe in-memory).
Explicit boundaries enforced by code (no direct DB table access from outside the owning module).

Goal: keep modules logically separated so extracting them later is easier.

Testing & CI for modular monoliths

Unit tests per module.
Integration tests that exercise module interactions.
Contract tests if you plan to extract a module later (define its API expectations).
Single CI pipeline with module-aware steps:
- Run unit tests for changed modules only.
- Run full integration test suite on main branch or PRs that touch multiple modules.
Keep migration tests (if extracting DB) to ensure data integrity during extraction.

Data and transactions

Monolith: ACID transactions are easy — same DB, same transaction boundary.
Microservices: distributed transactions are hard. Typical solutions:
- Eventual consistency with compensating actions (sagas).
- Domain events and event-driven propagation.
- Using orchestration or choreography patterns.

Plan early for how data will be owned and migrated if you extract modules later.

Migration strategy: how to go from modular monolith → microservices

Design modularly up front — well-defined modules, public APIs, and tests.
Measure pain points — slow deploys, test/compile times, team bottlenecks, or scaling issues.
Pick a candidate module — choose a module with stable boundaries and low coupling.
Extract its API — make sure the module already has an explicit API and tests.
Separate the datastore (if needed) — migrate data ownership carefully with an anti-corruption layer.
Create independent CI/CD for the extracted service.
Introduce monitoring & tracing for the new service (from the first deploy).
Iterate — repeat for other modules only when the benefit is clear.

Decision checklist (use this before picking microservices)

Is the product direction stable or still exploring fundamental features?
Are domain boundaries well-understood and unlikely to change soon?
Are there multiple teams that need independent release cadence?
Do you have (or can you hire) devops/sre expertise to operate distributed systems?
Are you hitting scaling issues that cannot be solved by vertical scaling or caching?
Can you afford the extra infra and operational cost?

If you answer “no” to most of these, default to a modular monolith.

Metrics to watch (to justify extraction later)

Deploy frequency and failure rate
Lead time for changes (how long it takes to get a change into prod)
Mean time to recovery (MTTR)
Code churn in specific modules (hotspots)
Service latency and resource usage per module
Team blocking/queueing (e.g., PRs waiting on a single owner)

These metrics tell you whether architecture is the bottleneck or whether process/product issues are to blame.

Common anti-patterns & pitfalls

Premature microservices: splitting before you understand the domain.
Monolith without modularity: single repo with no structure; prevents later extraction.
One DB shared by multiple microservices with direct table access (creates coupling).
No observability from the start — debugging distributed systems without tracing and logs is painful.
Extracting services too early — leads to churn and high maintenance cost.

Example scenarios

Small startup (5 people building MVP): Modular monolith. One deploy, low cost, fast iteration.
E-commerce at scale with many teams: Microservices for checkout, catalog, inventory, and payments, if you have the ops and engineering muscle.
SaaS with uncertain product-market fit: Modular monolith until product-market fit; extract after you know which modules are stable and high-load.

Diagrams (ASCII)

Modular Monolith

+----------------------------------------+
| App (single deploy)                    |
|  +--------+  +--------+  +--------+    |
|  | Users  |  | Orders |  | Billing|    |
|  | Module |  | Module |  | Module |    |
|  +--------+  +--------+  +--------+    |
+----------------------------------------+

Microservices

+--------+   +--------+   +--------+
| Users  |   | Orders |   | Billing|
| Service|   | Service|   | Service|
+--------+   +--------+   +--------+
   |            |            |
   +---- API / Messaging ----+

Practical checklist to implement a modular monolith today

Define modules (bounded contexts) and responsibilities.
Enforce module boundaries in code (no cross-module internal access).
Create public interfaces / APIs for each module.
Add module-level unit tests + cross-module integration tests.
Implement an in-process event bus for decoupling.
Keep a single CI pipeline with module-aware steps.
Add logging, metrics, and basic tracing from day one.
Document APIs and expected contracts (helps future extraction).

Closing thoughts

Microservices are not inherently “better” — they’re a different trade-off. For small teams and early-stage products, the fastest path to learning and validating product assumptions is typically a well-structured modular monolith. Build modularly, measure, and extract only when the cost-benefit is clear.